Content uploaded by Elise Murowchick
Author content
All content in this area was uploaded by Elise Murowchick on Jul 25, 2020
Content may be subject to copyright.
Adolescents’ sexual risk-taking has been conceptualized
in various ways: early age at first intercourse, number of
partners, type of partner or length of relationship, frequency
of intercourse, consistency of condom use, and use of other
methods of birth control. Although each can be considered
an aspect of risk taking, none by itself is valid as an opera-
tionalization of risky behavior (Metzler, Noell, & Biglan,
1992; Sieving et al., 1997). Rather, each is a proxy—a mea-
sure that captures some of the variance in risk-taking but
does not in itself completely measure the construct.
The reason that none of these factors completely repre-
sents risk-taking is that sexual behavior always involves a
combination of them. The characteristics of this combina-
tion determine whether a particular pattern of behavior is
safe or risky with regard to disease prevention or pregnan-
cy. For example, the risk of STD acquisition from incon-
sistent condom use may be minimal with one partner, like
in a mutually monogamous relationship, if neither partner
is infected. However, risk would become greater with larg-
er partner numbers because the risk of coming across an
infected individual increases. Similarly, disease risk from
condom non-use increases with larger number of inter-
course occasions, especially if occurring with multiple
partners. In terms of pregnancy, each additional act of
unprotected intercourse increases risk.
The use of risk-taking proxy measures may account for
some inconsistent findings in studies of adolescent sexual
risk-taking. For example, studies examining condom use
and number of partners typically find that one, but not the
other, is related to STD acquisition. Sometimes condom
use, but not number of partners, is found to be a statisti-
cally significant predictor of getting an STD (Upchurch,
Brady, Reichart, & Hook, 1990). At other times, number of
partners, but not condom use, is a predictor (Joffe et al.,
1992). Studies that account for more complex behavioral
patterns may improve understanding of adolescent sexual
risk-taking. Additionally, information about these patterns
can inform development of interventions that are more
reflective of the options that adolescents are comfortable
adopting and more effective than simple prescriptions to
use condoms or have fewer partners.
Previous Research Using Combined Risk Variables
Researchers have used two approaches for creating com-
posite measures of risk-taking. Most often, they have used
investigator-developed criteria to categorize participants
(e.g., high vs. low risk) or developed scales representing
degree of risk (e.g., Capaldi, Stoolmiller, Clark, & Owen,
2002; Metzler et al., 1992; Millstein & Moscicki, 1995;
Sieving et al., 1997). In two studies (Capaldi et al., 2002;
Millstein & Moscicki), the resulting measures were indeed
correlated with STD acquisition, supporting the value of
this approach. However, a drawback of this approach is
that it does not necessarily provide descriptive information
on the nature of the behaviors engaged in by individuals at
the different risk levels. Also, because it is determined a
priori, it may obscure or overlook important differences
among subgroups of risk-takers.
In contrast to this approach, Newman and Zimmerman
(2000) used a statistical method, Cluster Analysis, to
derive risk profiles empirically in a sample of 15- to 18-
year-old African Americans who reported having had sex.
They identified four subgroups: low-risk youth who used
The Journal of Sex Research Volume 42, Number3, August 2005: pp. 192–202 192
Condom Use, Frequency of Sex, and Number of Partners: Multidimensional
Characterization of Adolescent Sexual Risk-Taking
Blair Beadnell, Diane M. Morrison, Anthony Wilsdon, Elizabeth A. Wells, Elise Murowchick, Marilyn Hoppe,
Mary Rogers Gillmore, and Deborah Nahom
University of Washington School of Social Work
Sexual health research often relies on single risk indicators. However, multi-variable composites may better capture the
underlying construct of risk-taking. Latent Profile Analysis identified subgroups based on condom use consistency, partner
numbers, and sex frequency among 605 adolescents. Three profiles were identified for each of grades 8 to 10 (Condom
Users, Few Partners, and Risk-Takers) and 4 in grades 11 and 12 (Condom Users, One Partner, Two Partners, and Risk-
Takers). Inconsistent condom use groups reported more non-condom (and often less effective) birth control use and STD and
pregnancy histories. Females had greater representation in the Few Partners, One Partners, and Two Partners groups,
which also contained increasing proportions of participants in each subsequent year. Males had greater representation in
the Risk-Takers group. A profile approach to measurement has methodological advantages, can add to substantive knowl-
edge, and can inform content, timing, and targets of sexual health interventions.
Note. Preparation of this article was facilitated by a research grant from the
National Institute of Mental Health (MH63274-01) and a grant from the National
Institute on Drug Abuse (DA07047).
Address correspondence to Blair Beadnell, University of Washington School
of Social Work, 4101 15th Avenue NE, Seattle, WA, 98105; e-mail:
blairb@u.washington.edu.
Beadnell, Morrison, Wilsdon, Wells, Murowchick, Hoppe, Gillmore, and Nahom 193
condoms consistently and had fewer partners, high-risk
youth who used condoms inconsistently with many part-
ners, “monogamy strategy” youth who used condoms
inconsistently but had few partners, and “condom strate-
gy” partners who used condoms consistently with a mod-
erate number of partners. This approach has two advan-
tages. First, without imposing possibly incorrect a priori
assumptions, the researchers could see how risk factors
combined in the sample. Second, the method is informa-
tive in that examination of the subgroups provides under-
standing about the nature of the risks being taken and how
membership is related to other variables. For example, the
higher risk subgroup contained more males and the
monogamy subgroup contained more females. High-risk
males were found to use more alcohol and drugs during
sexual activity than those in other subgroups.
Purpose and Hypotheses
In this study, condom use, frequency of sex, and number of
partners were chosen as variables from which to derive
empirical risk-taking profiles, or subgroups, in a sample of
adolescents interviewed annually from the 8th to 12th
grades. Each of these variables can be considered to reflect
an aspect of risk-taking. Condom use, the factor most com-
monly used in research as a risk indicator, has been shown
to reduce (but not necessarily eliminate) the risk of disease
acquisition and pregnancy (Stone, Grimes, & Magder,
1986). For STDs, high frequency of intercourse is a risk
factor because when unprotected, it offers increased
opportunities for disease transmission. Frequent change in
sexual partners is also an important component of risk-tak-
ing since it exponentially increases the possibility of con-
tacting someone who is infected (Padian, Hitchcock,
Fullilove, Kohlstadt, & Brunham, 1990).
This study tested two hypotheses concerning sexual risk
behavior. In Hypothesis 1, we expected that profiles could
be identified in each grade using a latent variable
approach. Theoretically, this methodology allows the iden-
tification of subgroups engaging in lesser versus greater
risk-taking, as well as a description of the source of risk for
each. Hypothesis 2 predicted that the profiles would be
related logically to other risk-taking variables. We chose
variables that other studies have shown to be associates of
risk-taking in adolescents: gender (Donald, Lucke, Dunne,
O’Toole, & Raphael, 1994; MacKellar et al., 2000;
Newman & Zimmerman, 2000), STD and pregnancy his-
tory (Crosby et al., 2002; Metzler et al., 1992; Millstein &
Moscicki, 1995), and use of birth control methods other
than condoms (Civic, 1999; Sieving et al., 1997).
In addition to these hypotheses, we planned to assess
whether Newman and Zimmerman’s (2000) results are
generalizable when examined on a more ethnically diverse
sample, and with a different statistical approach (Latent
Profile Analyses) and methodology (deriving separate pro-
files for each age group rather than pooling age groups).
Although the major focus of the study hypotheses was to
derive and validate risk-taking profiles, we expected that
the findings also would yield interesting patterns that fur-
ther illustrate the utility of this methodological approach.
For example, we expected that a larger proportion of stu-
dents would be seen with higher risk profiles in later grades
than in earlier grades, given that older adolescents are more
likely to have sex (Brooks-Gunn & Paikoff, 1997) and less
likely to use condoms (Grunbaum et al., 2002).
M
ETHOD
All study procedures, measures, and protocols were
reviewed and approved by the University of Washington’s
Human Subjects Review Board.
Sample
We conducted a longitudinal study of four cohorts of stu-
dents (in grades 3 through 6 during the first data collection
wave in 1992). Students in a large, urban northwest school
district were recruited and then surveyed annually for seven
years. To maximize geographic, socioeconomic, and
racial/ethnic diversity, schools were selected based on
information about percentage of free lunch-eligible stu-
dents, racial/ethnic profiles, and geographic locations.
Parents or guardians of the students received brochures that
described the study and were asked for written consent for
their children to participate. Atotal of 2,319 students were
enrolled in the eligible grade levels at these schools. After
removing 203 students who were non-English speakers,
had changed schools, or could not be found, we obtained
parental consent for 1,177 (56%) participants. This consent
rate is similar to those obtained in other school-based stud-
ies requiring active parental consent (e.g., DeLoye,
Henggeler, & Daniels, 1993). We then asked all children of
consenting parents to assent to study participation. Four
children of consenting parents refused to participate, result-
ing in an initial sample of 1,173 (a 55% consent rate for
children). When the study was extended beyond the three
years originally planned, we obtained re-consent for 1,084
(92% of the original sample).
The data reported in this study come from assessments
of the two oldest cohorts when they were in 8th through
11th grades and from the oldest cohort in 12th grade. We
did not include the younger two cohorts, who did not reach
later years of high school during the study period and thus
had lower rates of sexual activity, in the analyses. Out of
the 627 participants originally recruited into the two oldest
cohorts, 605 (96%) were included in the analyses, as each
provided data for at least one time point from grades 8
through 12 (n= 597, 558, 560, 555, and 294 for grades 8
through 12, respectively). Of these, 261 (43%) and 344
(57%) were from the third and fourth cohort, respectively.
See Table 1 for participant demographics.
Aslight selection bias is likely in this sample, but there
was no detectable bias resulting from attrition once
enrolled. The sample contains a lower percentage of ado-
lescents who were eligible for the federally funded or
reduced price lunch program (35%), compared to the
school district population (44%) for these grades, suggest-
194 Characterization of Adolescent Sexual Risk-Taking
ing bias toward higher-income families. However, the
sample’s distribution of race and gender were not signifi-
cantly different from the school district population in the
same grades. In terms of study attrition, the 605 individu-
als included did not differ in terms of gender, ethnicity, or
income from the overall sample of 627. Similarly, attrition
did not appear to bias the sample in later years: there were
no significant differences in gender, ethnicity, or income
between the sample included in these analyses and the
larger sample at study enrollment.
Research Assessments
We administered surveys at yearly intervals, in the spring,
to small groups of 1 to 25 participants. Prior to survey
administration, interviewers set up the classroom environ-
ment to prevent each student from being able to see other
students’ answers (e.g., arranging seats at a distance from
each other and providing screens, when necessary, to block
students’ views of each others’ responses). Trained inter-
viewers read the questions aloud as the youth read along
and recorded their answers in their own written copies;
reading the survey aloud mitigated problems for those stu-
dents who were not strong readers. Teachers were not pre-
sent during survey administration, nor were students not
taking the survey. The survey was designed so that each
question had an answer that could be circled so that other
students could not deduce behavior by observing whether
an answer was being circled. Administration took about 45
minutes. Surveys occurred primarily in the schools. In
years 2 through 7, we interviewed participants who had
moved out of the area individually or, in rare cases, by
phone. Each youth received a gift such as a t-shirt or gift
certificate each year for participating.
Analytic Approach
Latent Profile Analysis. Latent Profile Analysis (LPA) is
an extension of Latent Class Analyses (LCA) that is appro-
priate when the observed variables of interest are mea-
sured as continuous (Gibson, 1959; Lazarsfeld & Henry,
1968). Renewed interest in the use of this method has
occurred as high-speed computers make such computa-
tionally-intensive methods practically applicable. Vermunt
and Magidson (2002) have described the numerous labels
used to describe this type of modeling, such as mixture-
likelihood approach to clustering, mixture-model cluster-
ing, probabilistic clustering, Bayesian classification, and
latent class cluster analysis. LPA is similar in intent to
other clustering approaches, such as cluster analysis and
Meehl’s taxonomic procedures. Cleland, Routhschild, and
Haslam (2000) describe the similarities and differences
between these techniques and report on simulation studies
which found LPA’s accuracy to be superior to cluster
analysis but similar to Meehl’s taxonomic procedures.
Several authors have described the intent and estimation
of LPA models (e.g., Gibson, 1959; Lazarsfeld & Henry,
1968; Muthén & Muthén, 2000; Vermunt & Magidson,
2002). As a latent variable approach, LPAassumes that the
relationship between a set of indicator variables can be
explained by a latent variable that is categorical. The cate-
gories are groups of people (referred to as classes) who are
similar to each other but different from people in the other
classes. The derivation of the latent class variable is based
on the idea of local independence: the classes identified
should be homogenous enough that the indicators are inde-
pendent (i.e., uncorrelated) among members of any partic-
ular class. This approach is especially useful in deriving
“empirically-based typologies such as personality catego-
rization, psychiatric syndromes, job interest constellations,
or modes of political behavior” (Gibson, 1962, p. 399).
Although LPA bears similarity to another technique, factor
analysis, these differ in their objectives. Factor analysis
seeks to derive continuous latent variables that explain the
relationships among a set of observed variables, while
LPA seeks to derive one categorical latent variable repre-
senting groups of individuals (Muthén & Muthén, 2000).
This type of analysis is a model-based and probabilistic
approach. Because it is model-based, a statistical model is
postulated for the population from which the sample is
recruited, measures of model fit are provided, and analyses
can be exploratory or confirmatory (Vermunt & Magidson,
2002). Being probabilistic, LPA can take uncertainty about
class membership into account even though each person is
assumed to belong to one class. For each person, the proba-
bility of being in each class is computed based on model
estimates and subjects’ observed scores on indicators
(Vermunt & Magidson). Class membership for an individual
is then based on the highest probability class for that person.
In LPA, the researcher typically determines the number
of subgroups that exist in a sample by performing the
analyses repeatedly, each time specifying an increasing
number. The solutions are compared, and the one chosen is
that with the best fit to the data. The primary measure of fit
used here was the Bayesian Information Criterion value
(BIC), which balances two components of a model, the
likelihood value and parsimony (Muthén & Muthén,
2000). Lower BIC values typically reflect better fit to the
data, and reductions of 6 or greater are considered “strong”
Table 1. Demographic Characteristics of the Sample (n= 605)
Characteristic % n
Gender
Female 50% 304
Male 50% 301
Race
African American 19% 115
Asian American 21% 125
European American 47% 283
Other 14% 82
Family Income1
< $17,420 7% 43
$17,420 - $24,790 27% 165
> $24,790 66% 397
1Based on eligibility categories for the federal free lunch program.
Note. Percents may not equal 100% due to rounding.
Beadnell, Morrison, Wilsdon, Wells, Murowchick, Hoppe, Gillmore, and Nahom 195
and 10 or greater “very strong” evidence of fit improve-
ment (Raftery, 1995). In addition to BIC values, other fac-
tors for choosing the superior solution are the inter-
pretability of the results, theoretical meaningfulness of the
profiles, and the classification quality (Muthén & Muthén,
2000). The latter is reflected in the ability to distinguish
membership in the latent profile groups given the model
and the data, reflected in higher “average class probabili-
ties” (ability to accurately classify individuals into their
most likely subgroup).
Hypothesis testing. For Hypothesis 1, we performed
LPA to identify categories of participants based on the
three variables of interest: how many times sexual inter-
course had occurred, how consistently condoms were used
during acts of sexual intercourse, and the number of sexu-
al partners. Five sets of analyses were performed, each for
the entire sample when in a particular grade (8 through
12). The analyses were implemented using Mplus 2.02.
We tested Hypothesis 2 by comparing the subgroups
identified in the LPA on gender, use of birth control (other
than male condoms), and STD and pregnancy history. We
expected that differences between subgroups would estab-
lish evidence for or against their validity. The Chi square
statistic was used for all comparisons and, for each,
Cramer’s Vand V2were calculated as effect size estimates.
Cramer’s Vis a correlation coefficient, ranging from 0 to
1, based on the Chisquare statistic. Accordingly, Cramer’s
V2, as reported in the tables, is similar to r2and here
reflects how much of the variability in the dependent vari-
able is explained by subgroup membership. All analyses
were performed in SPSS 11.0.
Measures
The questionnaire addressed knowledge of and attitudes
about a variety of health-risk and health-promoting
behaviors. Beginning in 8th grade, we asked a range of
sexual history questions. To enhance reliability and valid-
ity, item development was informed by and tested in focus
groups with the target population, and questions were
designed to have direct and clear wording with simple
response categories. Specific items used in these analyses
are described below.
Sexual behavior. We asked students, “Since you’ve been
in [present] grade, have you had sexual intercourse?”
Sexual intercourse was defined as “vaginal or anal inter-
course between any two people. Vaginal intercourse means
the penis in the vagina. Anal intercourse means the penis
in the rectum or butt.” Responses were “no” or “yes,”
coded as 0 or 1.
Number of sexual partners. Students responded to the
question, “Since you’ve been in [present] grade, how
many people have you had sexual intercourse with?” by
filling in the number of people.
Number of acts of intercourse. Students responded to,
“Since you’ve been in [present] grade, how many times
have you had sexual intercourse?” by filling in the number
of times.
Consistency of condom use. We asked students, “Since
you’ve been in [present] grade, how often have you or the
person you were having sex with used a condom when you
have had sexual intercourse?” Response options ranged
from 1 (never) to 7 (every time). In previous research
(Morrison, Baker, & Gillmore, 1998), this item was found
to have good predictive validity.
Birth control use. We assessed birth control use by ask-
ing students to circle “yes” or “no” for each method in an
exhaustive list of birth control methods: “Birth control pill;
rhythm or natural family planning (having sex at a time of
the month when you think pregnancy is not likely); jelly,
foam or cream spermicides; contraceptive sponge; with-
drawal (pulling out); condoms; Norplant; female condom;
diaphragm with spermicide; Depo-Provera (the shot);
other.” The item stem asked youth to “circle yes if you
have used the method since you’ve been in [present]
grade. Otherwise, circle no. Please include methods used
by the person you were having sex with.” To avoid appear-
ing to endorse questionable methods, a note at the bottom
of the list said, “Some of the things in the last question
help to prevent pregnancy, and some don’t. Please talk
with your parents, teacher, or school nurse if you need
more information.”
For the analyses, we created three dichotomous (did not
use/used, coded as 0 or 1) variables: use of rhythm or nat-
ural family planning, withdrawal, and female-controlled
methods. We constructed the latter based on whether a par-
ticipant or a sexual partner used at least one method over
which females typically have more control: birth control
pill; jelly, foam, or cream spermicides; contraceptive
sponge; Norplant; female condom; diaphragm with sper-
micide; or Depo-Provera. Pills and Depo-Provera were the
methods most often used (reported by 21% and 11% of
respondents, respectively, in the 12th grade), followed by
douching (5%) and jellies (4%). No other female-con-
trolled method was used by more than 2% of the sample at
any grade.
STD history. We asked, “Have you ever been told by a
doctor or other health care provider that you had a sexual-
ly transmitted disease?” Responses were “no” or “yes,”
coded as 0 or 1.
Pregnancy history. Females only were asked, “Have
you ever been pregnant?” Responses were “no” or “yes,”
coded as 0 or 1.
R
ESULTS
Preliminary Analyses
We first performed analyses using LPA for each cohort
separately. Because results for the two cohorts were virtu-
ally identical, we combined them into one sample and re-
analyzed the data to produce what is reported here.
Identification of Profile Groups using LPA
We selected only the participants who reported having had
sex and then performed LPA for each of five grades (8
196 Characterization of Adolescent Sexual Risk-Taking
through 12). For each grade, we first specified a 1-group
model. Then we performed the analysis repeatedly, each
time specifying an additional group (i.e., a 2-group model,
then a 3-group model, and so on). For each model, we
examined BIC scores. No further models were estimated
once the BIC score for a model increased from that of the
previous model. In certain models, we found that we had to
fix some variances to 0 to achieve convergence, always in
cases in which within-group variance was close to 0 for a
specific variable. The smallest BIC values occurred for the
3-group models in grades 8, 9, and 10 (BIC = 977, 1171, and
1693, respectively), and for the 4-group model in grade 11
(BIC = 2027). Average class probabilities were above .90
for these solutions. Although the BIC value was lowest for
the 5-group solution in grade 12 (BIC = 1344), the quality
of classification in that model was substantially worse (two
average class probabilities were below .90) than in the 4-
group solution (in which all average class probabilities were
above .90). Therefore, for grade 12, we selected the 4-group
model (BIC = 1367) for subsequent analyses.
Tables 2 and 3 show the characteristics of members of
each latent profile group for each grade. The characteris-
tics of the three profiles identified for grades 8, 9, and 10
were markedly similar across grades. One profile (which
we named Condom Users) consisted of participants who
tended to use condoms consistently with approximately
one to two partners; this group also had the fewest occa-
sions of sex. We considered this group the lowest risk
because of its combination of consistent condom use and
small number of partners. Members of the two other
groups had greater risk-taking because of inconsistent con-
dom use either with one to two partners (Few Partners) or
with a larger number of partners (Risk-Takers). In addition
to a greater number of partners, Risk-Takers also had the
most occasions of sex. Of the four profile groups identified
in grades 11 and 12, two were like the Condom Users and
Risk-Takers found in grades 8 through 10. Based on an
examination of the means and standard deviations, two
other groups appeared to be variations of the Few Partners
group found in the earlier grades. Members of both groups
used condoms inconsistently, either with exactly one part-
ner (the group labeled One Partner) or with approximately
two partners (the group labeled Two Partners).
Distributions of Participants Across Groups
Table 4 shows, for each grade separately, the distribution
of participants across profile groups. For descriptive and
theoretical purposes, we have included participants who
did not have sex in the table as a separate group (No Sex)
whose status was already known. Greater proportions of
participants reported having sex in each subsequent grade.
Among these, the Condom Users group was the largest in
the first year (grade 8). Although the proportion of partic-
ipants in this group became slightly larger over the years,
it eventually became second in size to groups representing
inconsistent condom use with small numbers of partners
(the Few Partners group in grade 10 and the One and Two
Partners groups in grades 11 and 12). The Risk-Takers
group increased across grade levels but remained the
smallest subgroup in grades 8 through 11 and was tied for
the smallest in grade 12.
Associations of Profile Group Membership
We found statistically significant differences in the gender
makeup of groups in each grade; see Tables 5 and 6.
Although the gender balance switched back and forth for
Condom Users across grades, more consistent findings
were seen in the other groups. The No Sex subgroup had a
fairly equal balance of males and females. However, groups
in which condoms were used inconsistently with few part-
ners (the Few Partners group in grades 8 through 10 and the
One Partners and Two Partners groups in grades 11 and 12)
typically had more females. Conversely, males had greater
representation in the Risk-Takers group, except in grade 10,
where the gender balance was roughly equal.
Tables 7 and 8 show the proportion of participants in
each group, and in each grade, who reported using birth
control methods other than male condoms. For each grade,
only participants who had sex were included in the analy-
ses. Statistically significant differences were seen in each
grade for at least one method. In all such cases, a greater
proportion of inconsistent condom use groups reported
using the birth control method. This was true for with-
drawal in all grades, rhythm in grades 10 and 12 (but not
8, 9, and 11), and female-controlled methods in later
grades (11 and 12), but not in earlier ones.
Profile group membership in many instances was asso-
ciated with STD and pregnancy history for grades in which
these questions were asked (9 through 12). For this analy-
sis, we collapsed into one group those that were character-
Table 2. Risk-Taking Profiles: Means (and Standard
Deviations) of Latent Profiles in Grades 8
Through 10
Latent Profile Groupa,b
Condom Users Few Partners Risk-Takers
8th Grade (n= 88)
How often use condomsc7.0 (0.2) 3.1 (1.5) 4.8 (2.0)
Number of sex partners 1.4 (0.6) 1.5 (0.7) 8.4 (3.9)
Number of times had sex 2.6 (1.5) 3.5 (1.6) 7.5 (4.5)
9th Grade (n= 145)
How often use condoms
C
7.0 (0.0) 3.5 (2.1) 5.4 (1.6)
Number of sex partners 1.4 (0.6) 1.3 (0.5) 5.4 (2.8)
Number of times had sex 4.7 (4.3) 7.3 (5.4) 10.6 (4.6)
10th Grade (n= 196)
How often use condomsc7.0 (0.0) 3.7 (2.0) 4.6 (1.9)
Number of sex partners 1.5 (0.8) 1.2 (0.4) 5.4 (4.9)
Number of times had sex 3.8 (2.6) 12.1 (8.4) 16.1 (7.1)
aCondom Users: Fewer sexual occasions with a small number of part-
ners, using condoms consistently; Few Partners: Medium number of
sexual occasions with a small number of partners, using condoms
inconsistently; Risk-Takers: Larger number of sexual occasions with
multiple partners, using condoms inconsistently.
bLatent profile groups were derived through Latent Profile Analysis
that included only sexually active participants.
cCondom use was measured on a scale of 1 (never) through 7 (every time).
Beadnell, Morrison, Wilsdon, Wells, Murowchick, Hoppe, Gillmore, and Nahom 197
ized by inconsistent condom use; otherwise, there would
be multiple cells with expected frequencies less than 5.
Table 9 shows the number of participants in the consistent
condom use and inconsistent condom use groups who
reported ever having an STD or becoming pregnant. For
STD history, relationships were statistically significant,
with effect sizes ranging from small to medium in magni-
tude, for all grades except 12. When statistically signifi-
cant, the findings were notable: except for two persons in
grade 11, all participants who reported ever having an STD
were in the inconsistent condom use groups. Among
females, inconsistent condom use was associated with
pregnancy history reported in grades 11 and 12, with medi-
um effect sizes, but not in grade 10. Due to low statistical
power, results are not shown for grade 9.
D
ISCUSSION
We empirically identified sexual behavior profiles of
teenagers, each of which reflected a different type or level
of risk-taking, and compared the subgroups on several
variables. Many of the findings discussed here illustrate
the utility of the latent profile approach to risk characteri-
zation while also providing substantively meaningful
information. This study provides new information by iden-
tifying risk subgroups in teens at different ages, showing
the complex and at times non-linear nature of relationships
among adolescent sexual risk behaviors, validating and
allowing greater generalizability of previous research,
illustrating the utility of the profile approach to risk mea-
surement, and providing ideas for targeted interventions.
Three profiles were identified in each of grades 8
through 10. Of the students who reported having sex, the
lowest risk group (Condom Users) used condoms consis-
tently with a small number of partners. The other profile
subgroups represented higher risk because of inconsistent
condom use, either with a small number of partners (Few
Partners) or with multiple partners (Risk-Takers). In
grades 11 and 12, two profiles emerged (One Partner and
Two Partners) seeming to represent a greater differentia-
tion of individuals who would have been grouped togeth-
er in the Few Partners profile in earlier years.
It is important to note that the inconsistent condom use
subgroups typically had more instances of sex then the
consistent condom use subgroups. More frequent sex
would add to pregnancy risk since each act of unprotected
intercourse would be an opportunity for pregnancy to
occur. In terms of STDs, more frequent sex combined with
condom non-use would provide greater opportunities for
disease transmission, especially due to a particular partner
being infected (for any of the subgroups). This risk can be
compounded for subgroups with multiple partners (the
Two Partner and Risk-Taker subgroups) due to the greater
likelihood of having at least one of multiple partners be
infected.
Table 3. Risk-Taking Profiles: Means (and Standard Deviations) of Latent Profiles in Grades 11 and 12
Latent Profile Groupa, b
Condom Users One Partner Two Partners Risk-Takers
11th Grade (n = 257)
How often use condomsc7.0 (0.0) 3.7 (2.1) 3.7 (1.8) 4.7 (1.9)
Number of sex partners 1.6 (0.9) 1.0 (0.0) 2.3 (0.4) 7.4 (3.8)
Number of times had sex 6.8 (9.9) 30.0 (28.0) 24.0 (22.1) 34.3 (25.0)
(logged)d0.6 (0.4) 1.2 (0.5) 1.2 (0.5) 1.4 (0.4)
12th Grade (n= 164)
How often use condomsc7.0 (0.0) 3.4 (2.0) 3.2 (1.8) 4.2 (1.9)
Number of sex partners 1.3 (0.5) 1.0 (0.0) 2.0 (0.0) 4.7 (2.2)
Number of times had sex 7.0 (7.9) 32.2 (32.4) 49.1 (37.1) 40.3 (37.8)
(logged)d0.6 (0.5) 1.3 (0.5) 1.5 (0.5) 1.4 (0.5)
aCondom Users: Fewer sexual occasions with a small number of partners, using condoms consistently; One Partner: Larger number of sexual occa-
sions with one partner, using condoms inconsistently; Two Partners: Larger number of sexual occasions with approximately two partners, using con-
doms inconsistently; Risk-Takers: Larger number of sexual occasions with multiple partners, using condoms inconsistently.
bLatent profile groups were derived through Latent Profile Analysis that included only sexually active participants.
cCondom use was measured on a scale of 1 (never) through 7 (every time).
dLog transformed scores on “Numer of times had sex” were used in analyses due to skewed distributions.
Table 4. Percentage Distribution of Participants Across
Latent Profile Groups in Each Grade
Latent Profile Groupa
Condom Few Risk-
Grade Users Partners Takers No Sexb
8th (n= 597) 9% 3% 2% 85%
9th (n = 558) 12% 10% 4% 74%
10th (n= 560) 12% 17% 6% 65%
One Two
Partner Partners
11th (n= 555) 15% 18% 9% 5% 54%
12th (n = 294) 15% 23% 9% 9% 44%
aCondom Users: Fewer sexual occasions with a small number of part-
ners, using condoms consistently; Few Partners: Medium number of
sexual occasions with a small number of partners, using condoms
inconsistently; One Partner: Larger number of sexual occasions with
one partner, using condoms inconsistently; Two Partners: Larger num-
ber of sexual occasions with approximately two partners, using con-
doms inconsistently; Risk-Takers: Larger number of sexual occasions
with multiple partners, using condoms inconsistently; No sex: Did not
have sex in that grade.
bParticipants who were not sexually active were not included in Latent
Profile Analysis because their status was already known.
Note. Row percentages may not add to 100 due to rounding.
198 Characterization of Adolescent Sexual Risk-Taking
Validity of Profile Groupings
Evidence supported the validity of these profile groupings.
First, they were conceptually sensible in that they repre-
sented profiles that are reasonable, given existing theoret-
ical and empirical evidence. Second, the number and char-
acteristics of the profiles were markedly similar across the
years, and grade 11’s increase in the number of profile sub-
groups was replicated in grade 12. Third, the results were
markedly similar to those of Newman and Zimmerman
(2000) who, as previously mentioned, found four clusters
among their sample of 15- to 18-year-old African
Americans. Despite differences between these studies in
methodology and sample characteristics, the profiles they
identified were similar to those found in this study’s grades
11 and 12. This suggests that their findings are robust and
generalizable beyond African Americans.
The findings concerning gender, given their consistency
with previous research, also provided evidence of profile
subgroup validity. For example, females typically had
greater representation in the Few Partners group (grades 8
through 10) and the One Partner and Two Partner groups
(grades 11 and 12). The only exception was the equivalent
representation in the Two Partners group in grade 11. This
is consistent with research showing reports of less condom
use and fewer partners on the part of female teens (Donald
et al., 1994; Durbin et al., 1993; Grunbaum et al., 2002;
MacKellar et al., 2000). Similarly, the Risk-Takers group
had more males than females in all but one grade, consis-
tent with findings of greater partner numbers on the part of
adolescent males (MacKellar et al.). In addition to their
consistency with previous research on adolescent risk-tak-
ing, both of these findings concerning gender replicate the
findings from Newman and Zimmerman’s (2000) study.
Other findings are consistent with previous research and
provide support for the validity of the profile groups
derived in the analyses. The finding that inconsistent con-
dom use was associated with pregnancy and STD histories
is not only logical, but also similar to research on the pro-
tective effects of condom use (Morrison, 1985; Stone et
al., 1986; Upchurch et al., 1990). Predictable from what is
known about teens was that the groups characterized by
inconsistent condom use or few partners reported higher
use of birth control methods other than male condoms.
Members of these groups probably had greater concern
about pregnancy than disease transmission due to low per-
ceived risk for STD acquisition (for some, because of hav-
ing a steady partner). This would be consistent with previ-
ous research showing that adolescents’ condom use varies
Table 5. Percentage of Sexually Active Males and Females in Each Latent Profile Group, Grades 8 Through 10
Latent Profile Groupa
Condom Users Few Partners Risk-Takers No Sex χ2df=3 Cramer’s V2b
8th grade (n= 56) (n= 18) (n= 14) (n= 509) 17.5*** .03
Female (n= 299) 30% 78% 29% 52%
Male (n= 298) 70% 22% 71% 48%
9th grade (n= 68) (n= 54) (n = 23) (n= 413) 8.7* .02
Female (n= 286) 60% 65% 39% 49%
Male (n= 272) 40% 35% 61% 51%
10th grade (n= 69) (n= 94) (n= 33) (n= 364) 11.5** .02
Female (n= 290) 54% 67% 52% 48%
Male (n= 270) 46% 33% 49% 53%
aCondom Users: Fewer sexual occasions with a small number of partners, using condoms consistently; Few Partners: Medium number of sexual occa-
sions with a small number of partners, using condoms inconsistently; Risk-Takers: Larger number of sexual occasions with multiple partners, using
condoms inconsistently; No sex: Did not have sex in that grade.
bInterpretation of Cramer’s V2: .01 = small effect size, .06 = medium, .16 = large.
*p. < .05 **p. < .01 ***p. < .001
Table 6. Percentage of Sexually Active Males and Females in Each Latent Profile Group, Grades 11 and 12
Latent Profile Groupa
Gender Condom Users One Partner Two Partners Risk-Takers No Sex χ2df=4 Cramer’s V2b
11th grade (n= 84) (n= 98) (n= 48) (n= 27) (n= 298) 29.4*** .05
Female (n= 288) 37% 75% 50% 41% 50%
Male (n= 267) 63% 26% 50% 59% 50%
12th grade (n= 44) (n= 67) (n= 26) (n= 27) (n= 130) 10.5* .04
Female (n= 149) 55% 63% 58% 30% 46%
Male (n= 145) 46% 37% 42% 70% 54%
aCondom Users: Fewer sexual occasions with a small number of partners, using condoms consistently; One Partner: Larger number of sexual occa-
sions with one partner, using condoms inconsistently; Two Partners: Larger number of sexual occasions with approximately two partners, using con-
doms inconsistently; Risk-Takers: Larger number of sexual occasions with multiple partners, using condoms inconsistently; No sex: Did not have sex
in that grade.
bInterpretation of Cramer’s V2: .01 = small effect size, .06 = medium, .16 = large.
*p. < .05 ***p. < .001
Beadnell, Morrison, Wilsdon, Wells, Murowchick, Hoppe, Gillmore, and Nahom 199
as a function of whether their primary goal is STD/HIV
prevention or pregnancy prevention (Crosby et al., 2001),
and that the relationship between condom use and other
contraceptive use is related to type of partner, length of
relationship, and perception of risk (Kershaw, Niccolai,
Ethier, Lewis, & Ickovics, 2003; Ott, Alder, Millstein,
Tschann, & Ellen, 2002; Roye & Seals, 2001).
Aquestion arises as to why an additional profile group
was found in grades 11 and 12 as compared to earlier
grades. Two explanations seem most likely. One is that the
greater number of participants having sex in the later years
allowed statistical differentiation into a larger number of
profile groups. However, this is not as one might expect,
given that the grade 12 data is based on a smaller sample.
The more likely explanation is that developmental changes
occurred that involved greater solidifying of romantic rela-
tionships, making stable monogamy more possible. Table
4 shows that a greater proportion of participants in each
subsequent grade were in the groups with fewer partners
(3%, 10%, and 17% in the Few Partners group in grades 8
through 10; 27% and 32% in the One Partner and Two
Partner groups thereafter). This may reflect a transition
from initial “exploration of sex for sex’s sake toward more
intimacy and commitment” found in qualitative interviews
Table 7. Percentage of Sexually Active Participants in Each Latent Profile Group Using Each Non-Condom Birth Control
Method, Grades 8 Through 10
Latent Profile Groupa
Birth Control Methods Condom UsersbFew PartnersbRisk-Takersbχ2df=2 Cramer’s V2c
8th grade (n= 53) (n= 18) (n= 14)
Female-Controlled Methodsd48% 50% 43% 0.2 .00
Withdrawal 19% 33% 64% 11.5** .13
Rhythm 06% 22% 7% 4.5 .05
9th grade (n= 68) (n= 54) (n= 23)
Female-Controlled Methodsd47% 43% 61% 2.2 .01
Withdrawal 21% 54% 39% 14.1*** .10
Rhythm 07% 17% 09% 2.8 .02
10th grade (n= 69) (n= 94) (n= 33)
Female-Controlled Methodsd39% 56% 53% 5.1 .03
Withdrawal 12% 62% 82% 58.9*** .30
Rhythm 04% 16% 19% 6.5* .03
aCondom Users: Fewer sexual occasions with a small number of partners, using condoms consistently; Few Partners: Medium number of sexual occa-
sions with a small number of partners, using condoms inconsistently; Risk-Takers: Larger number of sexual occasions with multiple partners, using
condoms inconsistently.
bPercentages reflect the proportion of members of each latent profile group who use that birth control method. In a particular group, the sum of per-
centages for the three methods may be greater or less than 100 because some individuals used more than one method, and some no method at all.
cInterpretation of Cramer’s V2: .01 = small effect size, .06 = medium, .16 = large.
dIncludes birth control pill; jelly, foam, or cream spermicides; contraceptive sponge; Norplant; female condom; diaphragm with spermicide; and/or
Depo-Provera.
*p<. .05 ** p< .01 ***p. < .001
Table 8. Percentage of Sexually Active Participants in Each Latent Profile Group Using Each Non-Condom Birth Control
Method, Grades 11 and 12
Latent Profile Groupa
Birth Control Methods Condom UsersbOne PartnerbTwo PartnersbRisk-Takersbχ2df=3 Cramer’s V2c
11th grade (n= 84) (n= 98) (n= 48) (n= 27)
Female-Controlled Methodsd43% 61% 63% 70% 10.1* .04
Withdrawal 17% 59% 52% 63% 39.8*** .15
Rhythm 08% 18% 15% 15% 3.8 .01
12th grade (n= 44) (n= 67) (n= 26) (n= 27)
Female-Controlled Methodsd34% 61% 62% 67% 10.7* .07
Withdrawal 16% 64% 65% 44% 28.5*** .17
Rhythm 02% 18% 19% 07% 7.9* .05
aCondom Users: Fewer sexual occasions with a small number of partners, using condoms consistently; One Partner: Larger number of sexual occa-
sions with one partner, using condoms inconsistently; Two Partners: Larger number of sexual occasions with approximately two partners, using con-
doms inconsistently; Risk-Takers: Larger number of sexual occasions with multiple partners, using condoms inconsistently.
bPercentages reflect the proportion of members of each latent profile group who use that birth control method. In a particular group, the sum of per-
centages for the three methods may be greater or less than 100 because some individuals used more than one method, and some no method at all.
cInterpretation of Cramer’s V2: .01 = small effect size, .06 = medium, .16 = large.
dIncludes birth control pill; jelly, foam, or cream spermicides; contraceptive sponge; Norplant; female condom; diaphragm with spermicide; and/or
Depo-Provera.
*p<. .05 ***p. < .001
200 Characterization of Adolescent Sexual Risk-Taking
with adolescents (Brooks-Gunn & Paikoff, 1997) and is
consistent with findings of greater closeness, intimacy, and
commitment on the part of older adolescents (Adams,
Laursen, & Wilder, 2001).
Implications
Identifying classes of individuals with varying risk levels
allows for the creation of new research questions and
hypotheses. Pursuing these can deepen the understanding
of adolescent risk behavior and contribute to theory-build-
ing. For example, these analyses identified a group of
Risk-Takers, young people with more sex partners, higher
frequency of sex, and relatively low condom use. In some
grades, this group showed considerable reliance on with-
drawal as a birth control method and was disproportion-
ately composed of males. Researchers might try to identi-
fy this group’s cultural, intrapersonal, and interpersonal
influences. Another salient question is whether individuals
tend to remain in the same profile over time, and if not,
what factors are associated with transitions from lower to
higher risk, or vice versa. Such information can inform the
timing and content of intervention approaches.
Through its use of a profile approach, this study also
helps to resolve conflicting findings in the literature. For
example, some studies have found that teens less likely to
use condoms tend to have only one partner (e.g.,
MacKellar et al., 2000), while others have found them to
have higher numbers of partners (e.g., Diclemente et al.,
1992). This study suggests that both findings have validi-
ty: of the individuals who used condoms inconsistently,
some had few partners (the Few Partners, One Partner, and
Two Partners groups) and, though a smaller percent of the
sample, some had multiple partners (the Risk-Takers
group). Thus, the relationship between number of partners
and condom use was not linear. Person-centered analytic
techniques sometimes yield more informative findings
compared to variable-centered methods, especially those
based on an assumption of linear relationships.
As noted in the introduction, it is difficult to represent
sexual risk-taking using single behavioral measures,
because sexual activity always involves a combination of
them. However, in more commonly used regression mod-
els collinearity, or correlations between risk factors used as
predictors, may lead to misleading conclusions. As an
example, the bivariate findings of one study (Boyer et al.,
2000) differed from its regression findings. In bivariate
analyses, both condom use and number of lifetime partners
were related to STD history. However, only number of
lifetime partners, and not condom use, was related to STD
history when both were entered simultaneously in a regres-
sion analysis with demographic and psychosocial vari-
ables. In light of the present study’s finding that inconsis-
tent condom use and higher numbers of partners some-
times co-occur, it seems plausible that the role of condom
non-use may have been obscured due to its correlation
with other variables in the equation, when, in fact, it was
an important factor.
Finally, these results, and results from similar future
studies, can inform intervention development. For exam-
ple, given that full abstinence among all teens is unlikely,
it is useful to distinguish teens who have sex but whose
risk level is low (i.e., the Condom Users group) from those
whose behavior carries a higher level of risk. Particularly
high-risk subgroups can then be targeted for additional or
tailored interventions. In this study, Risk-Takers in some
grades relied on withdrawal as a contraceptive method and
might therefore benefit from emphasis on its inefficacy, in
addition to encouragement to increase condom use or
decrease partner numbers. Consistent condom users had
lower rates of female-controlled methods, especially in
older grades when the frequency of sex was increasing.
Frank discussions of the greater contraceptive protection
afforded by dual method use would be indicated for these
teens. Similarly, those who limited partners to one or a few
appeared to put greater reliance on the rhythm method. We
know, however, that young users of rhythm typically can-
Table 9. Percentage of Consistent and Inconsistent Condom Users Reporting Lifetime STD and Pregnancy
STD Pregnancya
Consistent Inconsistent Consistent Inconsistent
Grade Users Usersbχ2df=1 Cramer’s V2 c Users Usersbχ2df=1 Cramer’s V2 c
9th (n= 66) (n= 74) not tabledd
00% 12% 8.6** .06
10th (n= 69) (n= 126) (n= 37) (n= 79)
00% 11% 8.3** .04 16% 27% 1.5 .01
11th (n= 84) (n= 146) (n= 31) (n= 96)
02% 10% 4.3* .02 07% 33% 8.6** .07
12th (n= 44) (n= 93) (n= 24) (n= 57)
05% 08% 0.4 <.01 08% 28% 3.8* .05
Note. STD and pregnancy history items were not asked of 8th graders.
aPregnancy applies only to female participants.
bBecause of multiple cells with expected frequencies less than 5, latent profile groups characterized by inconsistent condom use were collapsed into
one group.
cInterpretation of Cramer’s V2: .01 = small effect size, .06 = medium, .16 = large.
dPregnancy history is not shown for 9th graders due to low statistical power due to a relatively small number of sexually active females.
* p. < .05 ** p. < .01
Beadnell, Morrison, Wilsdon, Wells, Murowchick, Hoppe, Gillmore, and Nahom 201
not correctly identify their “safe periods” (Morrison,
1985), suggesting the usefulness of increased education
about this technique’s lack of efficacy among teens.
Study Limitations
It is important to consider the characteristics of this study
when interpreting its results. It has the strength of being
based on teens representative of the local public school
population. Additionally, we did not detect bias due to
attrition. However, several limitations must be kept in
mind. First, the sample can only be generalized to the
urban geographic area in which it was collected. Second,
we did not gather information on, and so could not
account for, whether participants’ sexual partners were
already infected with an STD, an important factor for
judging risk, especially in steady relationships. Third, our
measures of STD acquisition and pregnancy were cumu-
lative, as opposed to solely covering the grade in which
they were reported, and relied completely on self-report.
Finally, while some conclusions can be drawn concerning
developmental issues, a fuller examination was beyond
the scope of this study.
Conclusion
The methodology used in this study is one of several
approaches for conceptualizing and identifying risk. As
seen here, a person-centered latent variable approach can
make unique contributions to research on sexuality. Future
uses of this technique might include other risk factors,
such as participant and partner STD status, and other
known or suspected partner risk factors. Both basic and
applied research can be enriched by taking the interrelated
aspects of sexual risk-taking into account.
R
EFERENCES
Adams, R. E., Laursen, B., & Wilder, D. (2001). Characteristics of closeness
in adolescent romantic relationships. Journal of Adolescence, 24,
353–363.
Boyer, C. B., Shafer, M. A., Wibbelsman, C. J., Seeberg, D., Teitle, E., &
Lovell, N. (2000). Assocations of sociodemographic, psychosocial, and
behavioral factors with sexual risk and sexually transmitted diseases in
teen clinic patients. Journal of Adolescent Health, 27, 102–111.
Brooks-Gunn, J., & Paikoff, R. (1997). Sexuality and developmental transi-
tions during adolescence. In J. Schulenberg, J. Maggs, & K. Hurrelmann
(Eds.), Health risks and developmental transitions during adolescence.
New York: Cambridge University Press.
Capaldi, D. M., Stoolmiller, M., Clark, S., & Owen, L. D. (2002).
Heterosexual risk behaviors in at-risk young men from early adolescence
to young adulthood: Prevalence, prediction and association with STD
contraction. Developmental Pyschology, 38, 394–406.
Civic, D. (1999). The association between characteristics of dating relation-
ships and condom use among heterosexual young adults. AIDS Education
and Prevention, 11, 343–352.
Cleland, C. M., Rothschild, L., & Haslam, N. (2000). Detecting latent taxa:
Monte carlo comparison of taxometric, mixture model, and clustering
procedures. Psychological Reports, 87, 37–47.
Crosby, R. A., DiClemente, R. J., Wingood, G. M., Sionean, C., Cobb, B. K.,
Harrington, K., et al. (2001). Correlates of using dual methods for sexu-
ally transmitted diseases and pregnancy prevention among high-risk
African American female teens. Journal of Adolescent Health, 28,
410–414.
Crosby, R. A., DiClemente, R. J., Wingood, G. M., Sionean, C., Harrington,
K., Davies, S. L., et al. (2002). Pregnant African-American teens are less
likely than their nonpregnant peers to use condoms. Preventive Medicine,
34, 524–528.
DeLoye, G. H., Henggeler, S. W., & Daniels, C. M. (1993). Developmental
and family correlates of children’s knowledge and attitudes regarding
AIDS. Journal of Pediatric Psychology, 18, 209–219.
DiClemente, R. J., Durbin, M., Siegel, D., Krasnovsky, F., Lazarus, N., &
Comacho, T. (1992). Determinants of condom use among junior high
school students in a minority, inner-city school district. Pediatrics, 89,
197–202.
Donald, M., Lucke, J., Dunne, M., O’Toole, B., & Raphael, B. (1994).
Determinants of condom use by Australian secondary school students.
Journal of Adolescent Health, 15, 503–510.
Durbin, M., DiClemente, R. J., Siegel, D., Krasnovsky, F., Lazarus, N., &
Camacho, T. (1993). Factors associated with multiple sex partners among
junior high school students. Journal of Adolescent Health, 14, 202–207.
Gibson, W. A. (1959). Three multivariate models: Factor analysis, latent
structure analysis, and latent profile analysis. Psychometrika, 24,
229–252.
Gibson, W. A. (1962). Class assignment in the latent profile model. Journal
of Applied Psychology, 46, 399–400.
Grunbaum J. A., Kann, L., Kinchen, S., Williams, B., Ross J. G., Lowry R.,
et al. (2002 June 28). Youth Risk Behavior Surveillance - United States,
2001. In Centers for Disease Control, Morbidity and Mortality Report,
51(SS–4), 1–64.
Joffe, G. P., Foxman, B., Schmidt, A. J., Farris, K., Carter, R. J., Neumann,
S., et al. (1992). Multiple partners and partner choice as risk factors for
sexually transmitted disease among female college students. Sexually
Transmitted Disease, 19, 272–278.
Kershaw, T. S., Niccolai, L. M., Ethier, K. A., Lewis, J. B., & Ickovics, J. R.
(2003). Perceived susceptibility to pregnancy and sexually transmitted
disease among pregnant and nonpregnant adolescents. Journal of
Community Psychology. 4, 419–434.
Lazarsfeld, P. F., & Henry, N. W. (1968). Latent structure analyses. Boston,
MA: Houghton Mifflin.
MacKellar, D. A., Valleroy, L. A., Hoffmann, J. P., Glebatis, D., Lalota, M.,
McFarland, W., et al. (2000). Gender differences in sexual behaviors and
factors associated with nonuse of condoms among homeless and runaway
youths. AIDS Education and Prevention, 12, 477–491.
Metzler, C. W., Noell, J., & Biglan, A. (1992). The validation of a construct
of high-risk sexual behavior in heterosexual adolescents. Journal of
Adolescent Research, 7, 233–249.
Millstein, S. G., & Moscicki, A. (1995). Sexually transmitted disease in
female adolescents: Effects of psychosocial factors and high risk behav-
iors. Journal of Adolescent Medicine, 17, 83–90.
Morrison, D. M. (1985). Adolescent contraceptive behavior: A review.
Psychological Bulletin, 98, 538–568.
Morrison, D. M., Baker, S. A., & Gillmore, M. R. (1998). Condom use
among high-risk heterosexual teens: a longitudinal analysis using the
Theory of Reasoned Action. Psychology and Health, 13, 207–222.
Muthén, B., & Muthén, L. K. (2000). Integrating person-centered and vari-
able-centered analyses: Growth mixture modeling with latent trajectory
classes. Alcoholism: Clinical and Experimental Research, 24, 882–891.
Newman, P. A., & Zimmerman, M. A. (2000). Gender differences in HIV-
related sexual risk behavior among urban African American youth: A
multivariate approach. AIDS Education and Prevention. 12, 308–25.
Ott, M. A., Adler, N. E., Millstein, S. G., Tschann, J. M., & Ellen, J. M. (2002).
The trade-off between hormonal contraceptives and condoms among ado-
lescents. Perspectives on Sexual and Reproductive Health, 34, 6–14.
Padian, N., Hitchcock, P., Fullilove, R., Kohlstadt, V., & Brunham, R.
(1990). Report on the NIAID Study Group. Part I: Issues in defining
behavioral risk factors and their distribution. Sexually Transmitted
Diseases, 17, 200–204.
Raftery, A. E. (1995). Bayesian model selection in social research.
Sociological Methodology, 25, 111–164.
Roye, F., & Seals, B. (2001). A qualitative assessment of condom use deci-
sions by female adolescents who use hormonal contraception. Journal of
the Association of Nurses in AIDS Care, 12(6), 78–87.
Sieving, R., Resnick, M. D., Bearinger, L., Remafedi, G., Taylor, B. A., &
Harmon, B. (1997). Cognitive and behavioral predictors of sexually
202 Characterization of Adolescent Sexual Risk-Taking
transmitted disease risk behavior among sexually active adolescents.
Archives of Pediatric and Adolescent Medicine, 151, 243–251.
Stone, K. M., Grimes, D. A., & Magder, L. S. (1986). Primary prevention of
sexually transmitted diseases. Journal of the American Medical
Association, 255, 1,763–1,766.
Upchurch, D. M., Brady, E. W., Reichart, C. A., & Hook, E. W. (1990).
Behavioral contributions to acquisition of gonorrhea in patients attending
an inner city sexually transmitted disease clinic. Journal of Infectious
Diseases, 61, 938–941.
Vermunt, J. K., & Magidson, J. (2002). Latent class cluster analysis. In J. A.
Hagenaars & A. L. McCutcheon (Eds.), Applied latent class analysis.
New York: Cambridge University Press.
Manuscript accepted October 12, 2004
